New energy equipment early warning and decision-making method based on multi-source knowledge base and storage medium

Abstract

The invention relates to the technical field of power system automation, artificial intelligence application and supply chain management, and discloses a new energy equipment early warning and decision-making method based on a multi-source knowledge base and a storage medium, and the method comprises the steps that a state monitoring and early warning module calculates and predicts the residual life according to real-time operation data, and generates a fault early warning signal; the operation and maintenance knowledge base construction and reasoning module analyzes the early warning signal by using a power field knowledge graph and a retrieval enhancement generation technology and generates a material list; the supply chain linkage decision module determines an optimal supply source based on the global virtual inventory pool and a total cost objective function; and the closed-loop feedback optimization module performs online correction on the prediction model, the knowledge graph and the supply chain parameters by using the truth value data. According to the invention, fault early warning is realized through a collaborative linkage mechanism, and material supply is directly driven. The shutdown loss is introduced, so that the operation and maintenance comprehensive cost is minimized. According to the invention, the continuous accuracy and self-adaptability of decision making are guaranteed by using closed-loop feedback.

Description

Technical Field

[0001] This invention relates to the fields of power system automation, artificial intelligence applications, and supply chain management, specifically to a method and storage medium for early warning and decision-making of new energy equipment based on a multi-source knowledge base. Background Technology

[0002] New energy power generation equipment is typically deployed in geographically remote and harsh environments, and its operational stability directly impacts the economic benefits of power generation companies. To ensure the continuous operation of this equipment, maintenance personnel need to monitor its status in real time and promptly replace damaged components in case of malfunctions.

[0003] Existing operation and maintenance management models for new energy equipment typically separate fault early warning from material supply management into two independent processes. The fault early warning system focuses on predicting equipment status using sensor data, while the supply chain management system only passively responds to material needs after a maintenance work order is generated. This separation prevents fault early warning information from directly triggering the material preparation process, causing the allocation of spare parts to lag behind the moment the fault occurs, resulting in prolonged unplanned downtime of new energy power generation equipment.

[0004] Existing fault diagnosis technologies primarily rely on human experience or pre-set static rule bases for judgment. Faced with the massive amounts of unstructured logs and complex fault phenomena generated during the operation of new energy power generation equipment, static rule bases struggle to achieve accurate semantic parsing and fault location. Furthermore, existing technologies lack dynamic reasoning capabilities when establishing the mapping relationship between fault codes and required spare parts, easily leading to inaccurate mapping and resulting in incorrect or missed spare parts requisitions, thereby reducing on-site maintenance efficiency.

[0005] Existing material supply decisions are typically limited to the physical inventory of a single site or a fixed regional central warehouse. These decisions lack the capacity for coordinated allocation of inventory resources across regions and levels. Furthermore, existing material supply decisions often only consider the direct costs of logistics and transportation, failing to factor in the power generation losses caused by the shutdown of renewable energy power generation equipment due to waiting for supplies, thus failing to find the optimal balance between logistics costs and downtime losses.

[0006] Existing predictive models and supply chain parameters are typically determined through offline training during system deployment and remain static during subsequent operation. However, as the service life of new energy power generation equipment increases and the external supply chain environment fluctuates, the predictive accuracy of static models gradually decreases, and the preset supply chain parameters deviate from actual performance. Current technologies lack a mechanism for closed-loop correction of models and parameters using ground truth data generated during actual maintenance, resulting in the system's inability to adapt to the impacts of equipment aging and environmental changes.

[0007] Therefore, this invention proposes a method and storage medium for early warning and decision-making of new energy equipment based on a multi-source knowledge base to address the shortcomings of existing technologies. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention provides a method and storage medium for early warning and decision-making of new energy equipment based on a multi-source knowledge base. This solves the problem that existing technology prediction models and supply chain parameters cannot be closed-loop corrected using true data, resulting in their inability to adapt to the aging of new energy power generation equipment and changes in the external environment.

[0009] To achieve the above objectives, the present invention provides the following technical solution: This invention provides a method for early warning and decision-making of new energy equipment based on a multi-source knowledge base, comprising the following steps: S1. The status monitoring and early warning module collects real-time operating data of the new energy equipment. The status monitoring and early warning module uses a time-series data prediction model to calculate the predicted residual life of the new energy equipment based on the real-time operating data. When the predicted residual life is lower than a preset safety threshold, the status monitoring and early warning module generates a fault early warning signal. S2. The operation and maintenance knowledge base construction and reasoning module receives the fault warning signal, and the operation and maintenance knowledge base construction and reasoning module uses the power field knowledge graph to analyze the fault warning signal and generate an intelligent maintenance suggestion. The operation and maintenance knowledge base construction and reasoning module extracts a set of material lists from the intelligent maintenance suggestion. S3. The supply chain linkage decision module receives the material list set and the predicted residual life. The supply chain linkage decision module determines the supply response time of the warehouse node based on the global virtual inventory pool. The supply chain linkage decision module constructs a total cost objective function based on logistics costs and downtime losses. The supply chain linkage decision module solves the total cost objective function to determine the optimal supply source and generates a spare parts supply execution order. S4. The closed-loop feedback optimization module collects the true value data in the maintenance closure form, uses the true value data to perform online correction operation on the time series data prediction model, uses the true value data to perform dynamic update operation on the power field knowledge graph, and uses the true value data to perform adaptive calibration operation on the supply chain parameters.

[0010] Furthermore, when calculating the predicted residual life of new energy equipment, the condition monitoring and early warning module performs data cleaning processing on the real-time operating data. The module maps the collected real-time operating data into equipment condition monitoring vectors. It then selects these vectors from multiple consecutive time points and constructs an input sequence matrix. This input sequence matrix is ​​then fed into a time-series data prediction model, which outputs the predicted residual life through feature extraction and regression analysis.

[0011] Furthermore, when the O&M knowledge base construction and reasoning module generates intelligent maintenance recommendations, it transforms the fault codes and abnormal feature vectors contained in the fault warning signals into query vectors. The module then uses retrieval enhancement generation technology to retrieve the knowledge fragments with the highest similarity to the query vectors within the power industry knowledge graph. Finally, the module inputs the knowledge fragments and fault warning signals into a large language model. Using this large language model, the module combines the procedural content in the knowledge fragments with the abnormal features in the fault warning signals to infer and generate intelligent maintenance recommendations.

[0012] Furthermore, when constructing the power domain knowledge graph, the O&M knowledge base construction and reasoning module extracts key entities from unstructured and semi-structured data. These key entities are categorized into equipment objects, fault phenomena, fault causes, maintenance actions, required materials, and required tools. The O&M knowledge base construction and reasoning module defines a formal representation of the power domain knowledge graph, which consists of entity sets, relation sets, and triple sets. Within the power domain knowledge graph, the O&M knowledge base construction and reasoning module constructs a core fault handling chain, which sequentially connects fault code entities, fault phenomenon entities, component failure entities, maintenance strategy entities, and spare parts entities.

[0013] Furthermore, when determining the supply response time of a warehouse node, the supply chain linkage decision-making module constructs a global virtual inventory pool comprising four inventory levels: the local physical warehouse, neighboring warehouses, a regional central warehouse, and supplier-managed inventory and contracted inventory. The supply chain linkage decision-making module obtains the logistics time from the warehouse node to the site where the new energy equipment is located, as well as the procurement cycle corresponding to the material code. The supply chain linkage decision-making module also obtains the current available inventory level of the warehouse node. When the current available inventory level of the warehouse node is greater than zero, the supply chain linkage decision-making module determines that the supply response time equals the logistics time. When the current available inventory level of the warehouse node is zero, the supply chain linkage decision-making module determines that the supply response time equals the sum of the procurement cycle and the logistics time.

[0014] Furthermore, when constructing the total cost objective function, the supply chain linkage decision-making module defines the predicted residual life as the safety window. The module acquires the logistics costs incurred in transferring materials from the warehouse node to the site where the new energy equipment is located. It calculates the power generation loss cost per unit time of downtime for the new energy equipment. The module calculates the duration for which the supply response time exceeds the safety window and defines this duration as the unplanned downtime. Finally, the module defines the total cost objective function as the sum of logistics costs and downtime loss costs, where the downtime loss cost equals the power generation loss cost multiplied by the unplanned downtime.

[0015] Furthermore, when determining the optimal supply source, the supply chain linkage decision-making module iterates through all warehouse nodes in the global virtual inventory pool and calculates the total cost objective function value for each warehouse node. The module selects the warehouse node that minimizes the total cost objective function value as the optimal supply source. When there is no available inventory in the global virtual inventory pool, the module compares the safety window with a preset emergency procurement threshold. If the safety window is less than the emergency procurement threshold, the module executes emergency mode, prioritizing the supplier with the earliest expected delivery time. If the safety window is greater than the emergency procurement threshold, the module executes planning mode, using the economic order quantity model to calculate the procurement quantity.

[0016] Furthermore, when collecting truth data, the closed-loop feedback optimization module extracts the physical moment when the new energy equipment actually shut down or failed, defining this physical moment as the actual fault moment. It also extracts the material codes and quantities of spare parts actually replaced on-site by maintenance personnel, defining these as actual material consumption data. Finally, it extracts the timestamps of the actual delivery of spare parts to the site where the new energy equipment is located, defining these timestamps as the actual arrival moment. Finally, it extracts the generation time of the spare parts supply execution order from the external supply chain management system, defining this generation time as the order initiation time.

[0017] Furthermore, when the closed-loop feedback optimization module performs online correction operations, it calculates the absolute value of the difference between the actual failure time and the corresponding predicted residual lifetime time, defining this absolute value as the prediction bias. The closed-loop feedback optimization module compares the prediction bias with a preset allowable error threshold. When the prediction bias exceeds the allowable error threshold, the closed-loop feedback optimization module constructs corrected training sample pairs using the corresponding input sequence matrix and the actual failure time. The closed-loop feedback optimization module transmits the corrected training sample pairs to the time-series data prediction model and uses the backpropagation algorithm to perform incremental training operations to adjust the network parameters of the time-series data prediction model.

[0018] Furthermore, when the closed-loop feedback optimization module performs dynamic update operations, it locates the fault code entity corresponding to the actual fault code and the spare parts entity corresponding to the actual material consumption data in the power sector knowledge graph. The module then determines whether a direct material dependency relationship exists between the fault code entity and the spare parts entity. If no dependency relationship exists, the module creates a new triple in the power sector knowledge graph and uses the dependency relationship as a connection edge. If a dependency relationship already exists, the module updates the association weight value between the fault code entity and the spare parts entity using a positive incentive algorithm. When performing supply chain parameter adaptive calibration operations, the module calculates the difference between the actual arrival time and the order initiation time to obtain the actual delivery cycle. The module introduces a supplier delivery date correction coefficient and uses an exponentially weighted moving average algorithm to update the supplier delivery date correction coefficient based on the actual delivery cycle and the pre-stored procurement cycle. The closed-loop feedback optimization module uses the updated supplier delivery time correction coefficient to dynamically adjust the procurement cycle. The adjusted procurement cycle is then used by the supply chain linkage decision-making module to calculate the supply response time at subsequent moments.

[0019] The present invention also provides an electronic device, which includes a processor and a memory for storing processor-executable instructions, the processor being configured to execute the above-described new energy equipment early warning and decision-making method based on a multi-source knowledge base.

[0020] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-mentioned early warning and decision-making method for new energy equipment based on a multi-source knowledge base.

[0021] This invention provides a method for early warning and decision-making for new energy equipment based on a multi-source knowledge base. It has the following beneficial effects: 1. This invention utilizes a knowledge base construction and reasoning module to retrieve knowledge fragments from a power industry knowledge graph using retrieval enhancement generation technology. These knowledge fragments, along with fault warning signals, are then input into a large language model to generate intelligent maintenance recommendations. This solution addresses the difficulty of traditional rule-matching methods in handling unstructured fault descriptions. The knowledge base construction and reasoning module can accurately extract material lists from intelligent maintenance recommendations, improving the accuracy of fault diagnosis and material mapping.

[0022] 2. This invention constructs a global virtual inventory pool with four inventory levels through a supply chain linkage decision-making module, and builds a total cost objective function based on logistics costs and downtime losses. By solving the total cost objective function, the supply chain linkage decision-making module can accurately balance material allocation costs and power generation losses caused by the downtime of new energy equipment. Under the premise of ensuring that the supply response time meets the safety window, the supply chain linkage decision-making module selects the optimal supply source with the lowest overall economic cost, achieving dual optimization of operation and maintenance costs and downtime risks.

[0023] 3. This invention collects true data from maintenance closure reports through a closed-loop feedback optimization module, and uses this true data to perform online correction and dynamic updates to the time-series data prediction model, the power sector knowledge graph, and supply chain parameters. This closed-loop feedback mechanism eliminates the accuracy drift problem caused by static models over time. The closed-loop feedback optimization module enables the fault warning accuracy, knowledge graph association weights, and supplier delivery date correction coefficients to continuously adapt to the actual degradation characteristics of new energy equipment and the dynamic changes in the external supply chain, ensuring the long-term operational reliability of the system. Attached Figure Description

[0024] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a system block diagram of the present invention. Detailed Implementation

[0025] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] Please see Figure 1-2 This invention provides a method for early warning and decision-making of new energy equipment based on a multi-source knowledge base, including the following steps: S1. The status monitoring and early warning module collects real-time operating data of new energy equipment. The status monitoring and early warning module uses a time-series data prediction model to calculate the predicted residual life of new energy equipment based on real-time operating data. When the predicted residual life is lower than the preset safety threshold, the status monitoring and early warning module generates a fault early warning signal. S2. The Operation and Maintenance Knowledge Base Construction and Reasoning Module receives fault warning signals. The Operation and Maintenance Knowledge Base Construction and Reasoning Module uses the power field knowledge graph to analyze the fault warning signals and generate intelligent maintenance suggestions. The Operation and Maintenance Knowledge Base Construction and Reasoning Module extracts a set of material lists from the intelligent maintenance suggestions. S3, the supply chain linkage decision module receives the material list set and the predicted residual life. The supply chain linkage decision module determines the supply response time of the warehouse node based on the global virtual inventory pool. The supply chain linkage decision module constructs the total cost objective function based on logistics costs and downtime losses. The supply chain linkage decision module solves the total cost objective function to determine the optimal supply source and generates a spare parts supply execution order. S4. The closed-loop feedback optimization module collects the true value data from the maintenance closure form. The closed-loop feedback optimization module uses the true value data to perform online correction operations on the time series data prediction model. The closed-loop feedback optimization module uses the true value data to perform dynamic update operations on the power field knowledge graph and uses the true value data to perform adaptive calibration operations on the supply chain parameters.

[0027] See attached document Figure 2 This invention provides a new energy equipment early warning and supply chain linkage decision-making system based on a multi-source knowledge base. This system is applied in a physical environment including new energy power plants, centralized control centers, and multi-level warehousing networks. The system includes a status monitoring and early warning module, an operation and maintenance knowledge base construction and reasoning module, a supply chain linkage decision-making module, and a closed-loop feedback optimization module.

[0028] The condition monitoring and early warning module connects to the new energy intelligent operation and maintenance platform via an application programming interface (API). This module acquires real-time operating data from new energy equipment, including wind turbine generators and photovoltaic (PV) generators. The real-time operating data is high-frequency time-series data. For wind turbine generators, this data includes generator bearing temperature, gearbox oil temperature, blade angle, and vibration spectrum. For PV generators, this data includes string current, inverter conversion efficiency, ambient irradiance, and module temperature. The module performs data cleaning and noise reduction on the real-time operating data. Internally, the module deploys a time-series data prediction model. This model employs a long short-term memory (LSTM) network architecture or a Transformer architecture. The module uses this model to calculate the remaining lifetime of the new energy equipment. When the remaining lifetime falls below a preset safety threshold, the module generates a fault warning signal. This signal includes a fault code and an anomaly feature vector.

[0029] The Operations and Maintenance (O&M) Knowledge Base Construction and Reasoning module is connected to the Status Monitoring and Early Warning module. This module is used to construct a knowledge graph for the power industry and generate maintenance strategies. It includes a knowledge extraction unit that uses Optical Character Recognition (OCR) and Natural Language Processing (NLP) technologies to process unstructured data. This unstructured data includes O&M manuals, maintenance procedures, and fault code lookup tables. The knowledge extraction unit extracts entities and relationships from the unstructured data. Entities include fault code entities, fault phenomenon entities, maintenance action entities, and required material entities. Relationships include association relationships, cause relationships, solution relationships, and dependent material relationships. The O&M Knowledge Base Construction and Reasoning module uses these entities and relationships to construct a knowledge graph for the power industry. It also receives fault warning signals from the Status Monitoring and Early Warning module. The module uses retrieval enhancement generation technology and a large language model to parse these fault warning signals. Finally, it retrieves knowledge fragments from the power industry knowledge graph that match fault codes and anomaly feature vectors. The operation and maintenance knowledge base construction and reasoning module outputs structured maintenance strategies and material lists. These structured strategies and material lists include fault diagnosis results, recommended operating procedures, core material requirements, and auxiliary material requirements.

[0030] The supply chain linkage decision-making module connects the operations and maintenance knowledge base construction and reasoning module with the external supply chain management system. It is used to formulate spare parts supply strategies. The module constructs a global virtual inventory pool, which maps to the local physical warehouse, neighboring warehouses, regional central warehouses, and supplier-managed inventory. Based on the material list, the module queries the available inventory corresponding to the material codes in the global virtual inventory pool. It also obtains logistics time data and procurement cycle data from the external supply chain management system. Finally, the module calculates the total cost objective function. .

[0031] Total cost objective function The expression is: ; in, Represents the first in the global virtual inventory pool One warehouse; Indicates from the first The logistics costs incurred by a warehouse in transferring materials to the site where new energy equipment is located; This represents the cost of power generation loss per unit time incurred by new energy equipment during downtime; Indicates the first The warehouse will provide the supply response time required to deliver materials to the sites where new energy equipment is located; This indicates the remaining lifespan of the new energy equipment predicted by the condition monitoring and early warning module. This indicates the duration of unplanned downtime caused by the delivery of supplies being later than the remaining lifespan.

[0032] The supply chain linkage decision-making module iterates through all warehouses in the global virtual inventory pool. The supply chain linkage decision-making module selects the warehouses that maximize the total cost objective function. The smallest warehouse is selected as the optimal supply source. The supply chain linkage decision module generates a spare parts supply execution order. The spare parts supply execution order includes supply source information, estimated arrival time information, and logistics tracking number information.

[0033] The closed-loop feedback optimization module connects the supply chain linkage decision-making module, the status monitoring and early warning module, and the operation and maintenance knowledge base construction and reasoning module. The closed-loop feedback optimization module is used to correct system parameters. It collects real data from maintenance closure reports. This real data includes fault truth values, material consumption truth values, and time-to-delivery truth values. The module calculates the prediction deviation between the fault truth value and the remaining lifespan. It incrementally trains the time-series data prediction model based on the prediction deviation. It updates the association weights between entities in the power sector knowledge graph based on material consumption truth values. Finally, it updates the average supplier delivery time parameter in the external supply chain management system based on the time-to-delivery truth values.

[0034] This embodiment performs the steps of acquiring and preprocessing multi-source heterogeneous data. The status monitoring and early warning module connects to the group's new energy intelligent operation and maintenance platform through an application programming interface. The status monitoring and early warning module collects real-time operating data of new energy equipment at a set frequency.

[0035] For wind farms, real-time operating data includes generator bearing temperature, gearbox oil temperature, blade angle, and vibration spectrum. For photovoltaic (PV) farms, real-time operating data includes string current, inverter conversion efficiency, ambient irradiance, and module temperature.

[0036] The status monitoring and early warning module will be constantly The collected real-time operational data is mapped to device status monitoring vectors. Equipment status monitoring vector belong 3D real space . This represents the feature dimension, which corresponds to the number of physical quantity parameters collected.

[0037] The status monitoring and early warning module monitors equipment status vectors. Perform data cleaning operations. Data cleaning operations include interpolation and imputation of missing data and filtering and noise reduction of noisy data.

[0038] To capture the time-dependent characteristics of the operating status of new energy equipment, the status monitoring and early warning module constructs an input sequence matrix. Input sequence matrix It is a matrix composed of device status monitoring vectors from multiple consecutive time points. Input sequence matrix. The expression is: ; in: Indicates time The input sequence matrix; Indicates time The device status monitoring vector; The device status monitoring vector representing the start time of the time window; The device status monitoring vector represents the time window from the start time to the next time step. Indicates the length of the time window.

[0039] This embodiment performs a deep learning-based residual lifetime prediction step. The state monitoring and early warning module calls the internally deployed time-series data prediction model. The time-series data prediction model adopts a long short-term memory network architecture or a Transformer architecture. The state monitoring and early warning module will input the sequence matrix. Input into the time series data prediction model.

[0040] Time series data prediction models for input sequence matrices Feature extraction and regression analysis are performed. The condition monitoring and early warning module uses a time-series data prediction model to calculate the status of new energy equipment at time [time]. The predicted residual lifetime.

[0041] The formula for predicting residual lifetime is: ; in: Indicates at time The predicted residual lifespan of new energy equipment; A mapping function representing a time-series data prediction model; This represents the set of network parameters for a time-series data prediction model.

[0042] The status monitoring and early warning module has preset safety thresholds. The condition monitoring and early warning module will predict the remaining lifespan. With safety threshold Perform numerical comparisons.

[0043] When predicting residual lifetime Less than the safety threshold At that time, the status monitoring and early warning module determines that the new energy equipment has entered the fault early warning window period. The status monitoring and early warning module immediately generates a fault early warning signal.

[0044] Fault warning signals include fault codes and abnormal feature vectors Anomaly feature vectors For time series data prediction models at time... The hidden layer state vector at the last moment. The state monitoring and early warning module sends the fault early warning signal to the operation and maintenance knowledge base construction and reasoning module.

[0045] This embodiment executes the steps for constructing a knowledge graph in the power operation and maintenance (O&M) field. The O&M knowledge base construction and reasoning module is internally configured with a power field knowledge graph construction engine. This module utilizes the power field knowledge graph construction engine to process multi-source heterogeneous data.

[0046] Multi-source heterogeneous data is categorized into unstructured and semi-structured data. Unstructured data originates from documentation provided by equipment manufacturers. This documentation includes maintenance manuals, repair procedures, and fault code tables. Unstructured data is available in portable document formats and word processing document formats. Semi-structured data originates from historical maintenance work order records and material master data. Historical maintenance work order records contain text describing fault symptoms and text recording actual spare parts replacements. Material master data includes bill of materials (BOM) tables.

[0047] The Operations and Maintenance (O&M) knowledge base construction and reasoning module utilizes optical character recognition (OCR) technology combined with layout analysis algorithms to parse unstructured data. This module identifies spare parts lists and troubleshooting flowcharts in the O&M manual. Furthermore, it employs natural language processing (NLP) technology to extract key entities from unstructured and semi-structured data.

[0048] Key entities are categorized into different entity types. Entity types include equipment objects, fault phenomena, fault causes, maintenance actions, required materials, and required tools. The operation and maintenance knowledge base construction and reasoning module stores the extracted key entities and the relationships between them in a graph database. This module constructs a knowledge graph for the power industry.

[0049] The Operations and Maintenance Knowledge Base Construction and Reasoning module defines a formal representation of the knowledge graph in the power industry. The formal representation formula for the power industry knowledge graph is: ; in: Represents a knowledge graph in the power sector; Represents a set of entities in a knowledge graph of the power sector; Represents a set of relationships in a knowledge graph within the power sector; This represents the set of triples in the knowledge graph of the power sector.

[0050] Triple set Each triple in the graph consists of a head entity, a relation, and a tail entity. The operations and maintenance knowledge base construction and reasoning module constructs the core fault handling chain within the power industry knowledge graph. The core fault handling chain is represented as follows: ; in: This represents a fault code entity, which belongs to the entity set. ; The entity representing the fault phenomenon belongs to the entity set. ; This represents a component failure entity, belonging to the entity set. ; This represents the maintenance strategy entity, which belongs to the entity set. ; This represents a spare parts entity, belonging to the entity set. ; It represents an association relationship and belongs to a set of relations. , used to connect the fault code entity and the fault phenomenon entity; This indicates a cause-and-effect relationship and belongs to a set of relations. , used to connect the entity exhibiting the fault phenomenon with the entity exhibiting component failure; This represents a solution relationship and belongs to a set of relationships. , used to connect component failure entities and maintenance strategy entities; This indicates a dependency on resources and belongs to a set of relations. It is used to connect the maintenance strategy entity and the spare parts entity.

[0051] This embodiment executes the reasoning step based on retrieval enhancement. The operation and maintenance knowledge base construction and reasoning module receives fault warning signals from the status monitoring and early warning module. The fault warning signal contains a fault code and anomaly feature vector. The operation and maintenance knowledge base construction and reasoning module transforms the fault code and anomaly feature vector into a query vector. .

[0052] The Operations and Maintenance Knowledge Base Construction and Reasoning module performs vectorized retrieval operations within the power industry knowledge graph. This module calculates the query vector. With each entity vector in the knowledge graph of the power sector The cosine similarity.

[0053] The formula for calculating cosine similarity is: ; in: Represents the query vector With the entity vectors The cosine similarity value between them; This represents a query vector derived from a fault warning signal. Represents the first in the knowledge graph of the power sector One entity vector; Represents the query vector The model; Indicates the first entity vectors The model.

[0054] The operations and maintenance knowledge base construction and reasoning module sorts entity vectors based on their cosine similarity scores. The module selects the vectors with the highest cosine similarity scores. Each entity vector corresponds to a knowledge fragment. The knowledge fragment contains information about the faulty node and its neighboring nodes.

[0055] The O&M knowledge base construction and reasoning module calls the large language model. This module inputs selected knowledge fragments and original fault warning signals into the large language model. The large language model then performs the reasoning and generation task. Combining the procedural content from the knowledge fragments with the abnormal features in the fault warning signals, the large language model generates a structured intelligent maintenance suggestion.

[0056] The Operations and Maintenance Knowledge Base Construction and Reasoning module parses the intelligent maintenance suggestion. This module extracts a materials list from the intelligent maintenance suggestion. .

[0057] Material List Collection The expression is: ; in: This represents a collection of material lists; Indicates the first A unique material code for each type of material; Indicates the first The required quantity of various supplies; This indicates the total quantity of the required types of materials.

[0058] The operations and maintenance knowledge base construction and reasoning module outputs a structured data package containing confidence scores. This structured data package contains a set of material lists. The operations and maintenance knowledge base construction and reasoning module transmits structured data packages to the supply chain linkage decision-making module.

[0059] This embodiment executes the steps for constructing a global virtual inventory pool. The supply chain linkage decision-making module connects to an external supply chain management system via an application programming interface (API). The supply chain linkage decision-making module is no longer limited to querying inventory data from a single site. Instead, it constructs a global virtual inventory pool.

[0060] The global virtual inventory pool enables logical mapping of inventory data across different levels. The global virtual inventory pool comprises four inventory levels. The first level is the local site's physical warehouse, corresponding to the site where the faulty equipment is located. The second level is the neighboring site warehouses, corresponding to other sites in the same region or wind farm cluster. The third level is the regional central warehouse, corresponding to provincial or regional material centers. The fourth level is supplier-managed inventory and contracted inventory, corresponding to inventory held by external suppliers.

[0061] The supply chain linkage decision module defines the set of warehouses in the global virtual inventory pool. Warehouse Collection The expression is: ; in: This represents the collection of warehouses in the global virtual inventory pool; This represents the total number of warehouse nodes in the global virtual inventory pool.

[0062] The supply chain linkage decision-making module is based on the material list collection. The material codes in the database initiate real-time retrieval requests within the external supply chain management system. The supply chain linkage decision-making module obtains information from each warehouse node. The current available inventory level for the corresponding material code.

[0063] The supply chain linkage decision module defines the current available inventory as follows: . Indicates the first warehouse nodes The actual physical inventory quantity corresponding to the material code. The supply chain linkage decision module utilizes the current available inventory. Determine the warehouse node Does the system meet the conditions for direct allocation? The supply chain linkage decision-making module ensures that the data in the global virtual inventory pool is synchronized in real time with the physical inventory data in the external supply chain management system.

[0064] This embodiment performs the quantitative calculation steps for decision variables. The supply chain linkage decision module receives the prediction results from the condition monitoring and early warning module. The prediction results include the remaining lifespan of the new energy equipment.

[0065] The supply chain linkage decision-making module defines the remaining lifespan of new energy equipment as the safety window period. Safety window period The value is equal to the predicted residual life. The value.

[0066] The supply chain linkage decision-making module targets warehouse collections. Each warehouse node in Acquire logistics and transportation time data. The supply chain linkage decision-making module uses geographic information system data to calculate the time from the warehouse node. The physical transportation time to the site where the new energy equipment is located. The supply chain linkage decision module defines physical transportation time as logistics time. .

[0067] The supply chain linkage decision-making module acquires the procurement cycle data corresponding to the material codes. The procurement cycle data represents the delivery time required for suppliers to produce or ship materials. The supply chain linkage decision-making module defines the delivery cycle length as the procurement cycle. .

[0068] The supply chain linkage decision-making module combines current available inventory levels. Calculate each warehouse node The supply chain response time. The supply chain linkage decision module defines the supply response time as... .

[0069] Supply response time The calculation formula is: ; in: Represents a warehouse node Supply response time; Indicates from the repository node Logistics time to the station where the new energy equipment is located; Represents a warehouse node The current available inventory level; Indicates the supplier's procurement cycle; Represents a warehouse node There is spot inventory in the market; Represents a warehouse node There is no spot inventory in China.

[0070] The supply chain linkage decision-making module utilizes supply response time This represents the earliest time required for supplies to arrive on-site. The supply chain linkage decision-making module will consider supply response time. With safety window period Perform subsequent comparison calculations.

[0071] This embodiment executes the optimal supply strategy decision-making steps. The supply chain linkage decision-making module constructs a total cost objective function based on logistics costs and downtime losses. The supply chain linkage decision-making module obtains data from warehouse nodes. Logistics costs incurred in allocating materials to the sites where new energy equipment is located. The supply chain linkage decision-making module defines logistics costs as logistics expenses. .

[0072] The supply chain-linked decision-making module calculates the amount of power generation loss per unit time caused by the downtime of renewable energy equipment based on current wind speed or solar radiation forecast data. This module defines the amount of power generation loss as the power generation loss cost. .

[0073] The supply chain linkage decision-making module constructs a total cost objective function. This total cost objective function is used to evaluate the selection of warehouse nodes. The overall economic cost of being a source of supply.

[0074] The expression for the total cost objective function is: ; in: Indicates selecting a warehouse node. The total cost value at that time; Indicates from the repository node Logistics costs incurred from the allocation of supplies; This represents the cost of power generation loss per unit time incurred by new energy equipment during downtime; Represents a warehouse node Supply response time; Indicates the safe window period; Indicates the duration of unplanned downtime; This represents a zero value and is used to set the unplanned downtime to zero when the supply response time is less than or equal to the safety window.

[0075] The supply chain linkage decision module traverses the warehouse set. All warehouse nodes within the system. The supply chain linkage decision module calculates the value of each warehouse node. Corresponding total cost value The supply chain linkage decision-making module solves for the total cost value. Minimize the optimal supply source .

[0076] The supply chain linkage decision-making module executes hierarchical decision-making strategies based on the calculation results.

[0077] When at least one warehouse node exists This reduces supply response time Less than or equal to the safe window period At this time, the supply chain linkage decision module determines it to be a local fulfillment scenario or a regional allocation scenario. Unplanned downtime is zero at this point. Total cost value. Equal to logistics costs The supply chain linkage decision-making module prioritizes logistics costs. The lowest-ranking warehouse node is designated as the optimal supply source. The supply chain linkage decision-making module generates reservation orders or inter-site transfer proposals within the enterprise resource planning system of this station.

[0078] When all warehouse nodes Supply response time All are greater than the safe window period At this point, the supply chain linkage decision-making module determines the scenario as minimizing downtime losses. In this case, the unplanned downtime duration is greater than zero. The supply chain linkage decision-making module selects the total cost value. The smallest warehouse node is selected as the optimal supply source.

[0079] When there is no available inventory in the global virtual inventory pool, the supply chain linkage decision-making module triggers an external procurement strategy. The supply chain linkage decision-making module will then establish a safety window. Compare with the preset emergency procurement threshold.

[0080] If the safety window period If the demand falls below the preset emergency procurement threshold, the supply chain linkage decision-making module will execute emergency mode. This module will match the available inventory of the e-commerce platform or contracted suppliers. Ignoring price factors, the module will prioritize the supplier with the earliest estimated delivery time.

[0081] If the safety window period If the demand exceeds the preset emergency procurement threshold, the supply chain linkage decision-making module executes the planned procurement mode. The module places the material demand into a procurement pool. It then calculates the procurement quantity using the economic order quantity model. Finally, the module waits to merge similar demands from other sites before triggering a bulk procurement process.

[0082] The supply chain linkage decision-making module generates a spare parts supply execution order based on the determined optimal supply source. The spare parts supply execution order includes the supply source, estimated arrival time, and logistics tracking number. The supply chain linkage decision-making module then sends the spare parts supply execution order to an external supply chain management system for execution.

[0083] This embodiment performs the truth data feedback and verification steps. The closed-loop feedback optimization module connects to the external maintenance management system through an application programming interface. The closed-loop feedback optimization module monitors the execution status of maintenance work orders in real time.

[0084] When a maintenance work order's status changes to "closed," the closed-loop feedback optimization module automatically retrieves the maintenance closure form. The maintenance closure form is a record document filled out by maintenance personnel after completing on-site fault repair. The closed-loop feedback optimization module performs structured parsing on the maintenance closure form and extracts key truth data from it.

[0085] Key truth data includes fault truth data, material truth data, and timeliness truth data.

[0086] The closed-loop feedback optimization module extracts the physical moment when the new energy equipment actually shuts down or fails. The module defines this physical moment as the actual fault moment. Real-world failure moments Used to verify the prediction accuracy of the status monitoring and early warning module.

[0087] The closed-loop feedback optimization module extracts the material codes and quantities of spare parts actually replaced by maintenance personnel on-site. This module defines the material codes and quantities of the replaced spare parts as actual material consumption data. This actual material consumption data is used to verify the accuracy of recommendations made by the maintenance knowledge base construction and reasoning module.

[0088] The closed-loop feedback optimization module extracts the timestamp of the actual delivery of spare parts to the site where the new energy equipment is located. The closed-loop feedback optimization module defines the timestamp as the actual arrival time. Actual arrival time Used to verify the accuracy of logistics time estimation in the supply chain linkage decision module.

[0089] The closed-loop feedback optimization module will detect real-time faults. Actual material consumption data and actual delivery time The data is stored in the historical verification database. The closed-loop feedback optimization module uses the data in the historical verification database to initiate the subsequent parameter correction process.

[0090] This embodiment performs an online correction step for the predictive model. The closed-loop feedback optimization module reads the actual fault timestamps from the historical validation database. The closed-loop feedback optimization module calls the status monitoring and early warning module at any time. The generated historical forecast records. These records contain forecasts of remaining lifetime. .

[0091] The closed-loop feedback optimization module calculates the prediction deviation. The closed-loop feedback optimization module defines the prediction deviation as... .

[0092] Prediction bias The calculation formula is: ; in: Indicates time The absolute time difference between the predicted result and the actual result; Indicates the actual physical moment when a fault occurs in a new energy device; This indicates the moment when the status monitoring and early warning module performs the predictive operation; This indicates that the status monitoring and early warning module is at time... The output is the predicted residual lifetime; This indicates the predicted time of failure by the status monitoring and early warning module; This indicates the operation of taking the absolute value.

[0093] The closed-loop feedback optimization module has a preset allowable error threshold. The closed-loop feedback optimization module will predict the deviation. With the allowable error threshold Perform numerical comparisons.

[0094] When prediction deviation Greater than the allowable error threshold At that time, the closed-loop feedback optimization module determines that the time series data prediction model has accuracy drift. The closed-loop feedback optimization module extracts the time. The corresponding input sequence matrix The closed-loop feedback optimization module utilizes the input sequence matrix. and real failure moments Construct corrected training sample pairs.

[0095] The closed-loop feedback optimization module transmits the corrected training sample pairs to the state monitoring and early warning module. The state monitoring and early warning module then performs incremental training using the corrected training sample pairs. Finally, the state monitoring and early warning module uses the backpropagation algorithm to adjust the network parameter set of the time-series data prediction model. This process adapts the time-series data prediction model to the current degradation characteristics of the new energy equipment.

[0096] This embodiment performs an adaptive calibration step for supply chain parameters. The closed-loop feedback optimization module performs parameter calibration for the material supply process of fourth-tier supplier-managed inventory and contract inventory. The closed-loop feedback optimization module extracts the actual arrival time from the maintenance closure report. The closed-loop feedback optimization module extracts the generation time of the spare parts supply execution order from the external supply chain management system. The module defines the generation time of the spare parts supply execution order as the order initiation time. .

[0097] The closed-loop feedback optimization module calculates the supplier's actual delivery cycle. The actual delivery cycle equals the actual arrival time. Subtract the order initiation time The value. The closed-loop feedback optimization module reads the procurement cycle pre-stored in the external supply chain management system. Procurement cycle It is static data based on historical contractual agreements.

[0098] To eliminate the lag in static data and reflect the supplier's current fulfillment capability, the closed-loop feedback optimization module introduces a supplier delivery time correction factor. The closed-loop feedback optimization module uses an exponentially weighted moving average algorithm to update the supplier delivery date correction coefficient. .

[0099] Supplier delivery time correction factor The update formula is: ; in: This indicates the supplier delivery time correction factor after this update; This represents the smoothing factor, which is a real number between 0 and 1, used to adjust the sensitivity to recent data. Indicates the actual arrival time; Indicates the time the order was initiated; This indicates the procurement cycle pre-stored in the external supply chain management system; This represents the ratio of the actual delivery cycle to the pre-determined procurement cycle, used to characterize the degree of deviation in this supply. This indicates the supplier delivery time correction factor before this update.

[0100] The closed-loop feedback optimization module will update the supplier delivery time correction factor. Write it into an external supply chain management system.

[0101] When the supply chain linkage decision module is in a subsequent moment When performing the quantitative calculation steps of decision variables, the supply chain linkage decision module uses the updated supplier delivery date correction coefficient. Procurement cycle Perform dynamic corrections.

[0102] The dynamically adjusted procurement cycle calculation formula is as follows: ; in: This indicates the actual effective procurement cycle after dynamic adjustments; This indicates the procurement cycle pre-stored in the external supply chain management system; This indicates the updated supplier delivery time correction factor.

[0103] The supply chain linkage decision-making module utilizes the dynamically adjusted actual effective procurement cycle. Recalculate supply response time The supply chain linkage decision-making module ensures that the optimal supply strategy decision-making model is calculated based on the latest supply chain efficiency data.

[0104] This embodiment performs the dynamic update step of the knowledge graph. The closed-loop feedback optimization module connects to the operation and maintenance knowledge base construction and reasoning module. The closed-loop feedback optimization module calls the power field knowledge graph stored in the graph database.

[0105] The closed-loop feedback optimization module reads fault truth data and actual material consumption data from the historical verification database. The fault truth data includes the actual fault codes confirmed by maintenance personnel. The actual material consumption data includes the material codes of spare parts actually used by maintenance personnel on-site.

[0106] The closed-loop feedback optimization module locates the fault code entity corresponding to the actual fault code in the power industry knowledge graph. The closed-loop feedback optimization module also locates the spare parts entity corresponding to the material code in the power industry knowledge graph.

[0107] The closed-loop feedback optimization module determines whether there is a direct material dependency relationship between the fault code entity and the spare parts entity.

[0108] When there is no dependency relationship between the fault code entity and the spare parts entity, the closed-loop feedback optimization module determines that new operation and maintenance knowledge has emerged. The closed-loop feedback optimization module creates a new triple in the power domain knowledge graph. The new triple has the fault code entity as the head entity. The new triple has the spare parts entity as the tail entity. The new triple has the dependency relationship as the connecting edge. The closed-loop feedback optimization module writes the new triple into the graph database.

[0109] When a dependency relationship already exists between a fault code entity and a spare parts entity, the closed-loop feedback optimization module performs a value update operation on the association weight. The closed-loop feedback optimization module defines the association weight between the fault code entity and the spare parts entity as follows: Association weight It characterizes the probability that the spare part will be needed when a failure occurs.

[0110] The closed-loop feedback optimization module updates the associated weights using a positive incentive algorithm. .

[0111] Association weight The update formula is: ; in: This represents the updated association weight value; This indicates the current association weight value before the update; This represents the preset knowledge confidence learning rate, which is used to control the step size of weight growth. Indicates the upper limit of the association weight; Indicates the indicator factor for the use of supplies; The value is 1 when the actual material consumption data includes the entity of this spare part; The value is 0 when the actual material consumption data does not include the entity of the spare part.

[0112] The closed-loop feedback optimization module will update the associated weight values. Stored in a graph database. When the O&M knowledge base construction and reasoning module executes the reasoning step based on retrieval enhancement the next time, it uses the updated power domain knowledge graph for retrieval. The O&M knowledge base construction and reasoning module prioritizes recommending spare parts entities with higher association weight values.

[0113] This embodiment provides an electronic device. This electronic device is designed to execute the new energy equipment fault early warning, intelligent diagnosis, material mapping, and supply chain decision-making methods described in the above embodiments. The electronic device can be a server, workstation, personal computer, or cloud computing node.

[0114] At the hardware level, electronic devices mainly include processors, memory, communication interfaces, and communication buses.

[0115] The communication bus is used to connect and communicate between the processor, memory, and communication interface. The communication interface is used to enable data transmission between electronic devices and external network elements.

[0116] Memory can be high-speed random access memory or unstable memory, such as disk storage. Memory is used to store computer programs. Computer programs contain various program instructions.

[0117] A processor can be a central processing unit, an application-specific integrated circuit (ASIC), or one or more integrated circuits used to control program execution.

[0118] When the processor executes a computer program stored in memory, the processor is configured to perform the following steps: First, it performs the acquisition and preprocessing of multi-source heterogeneous data. The processor receives real-time running data and constructs an input sequence matrix. And perform data cleaning operations.

[0119] Second, perform residual lifetime prediction based on deep learning. The processor uses a time-series data prediction model to calculate the predicted residual lifetime. The processor will predict the remaining lifetime. With safety threshold The comparison is performed to generate a fault warning signal containing fault codes and abnormal feature vectors.

[0120] Third, it performs intelligent diagnosis and resource mapping based on knowledge graphs. The processor utilizes knowledge graphs in the power sector. and query vector Perform vectorized retrieval. The processor calculates cosine similarity. It also extracts knowledge fragments. The processor generates an intelligent maintenance suggestion and outputs a set of material lists. .

[0121] Fourth, execute supply chain coordination and dynamic decision-making. The processor constructs a global virtual inventory pool and calculates supply response time. The processor constructs and solves the total cost objective function. The processor determines the optimal supply source. And generate a spare parts supply execution order.

[0122] Fifth, implement closed-loop feedback and model self-evolution. The processor acquires the actual fault moments. Actual material consumption data and actual delivery time The processor calculates the prediction bias. The time-series data prediction model is incrementally trained. The processor updates the supplier delivery date correction coefficient. The procurement cycle is dynamically adjusted. The processor updates the association weights in the knowledge graph. .

[0123] Furthermore, this embodiment also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the processor implements a new energy equipment early warning and decision-making method based on a multi-source knowledge base.

Claims

1. A method for early warning and decision-making of new energy equipment based on a multi-source knowledge base, characterized in that, Includes the following steps: S1. The status monitoring and early warning module collects real-time operating data of the new energy equipment. The status monitoring and early warning module uses a time-series data prediction model to calculate the predicted residual life of the new energy equipment based on the real-time operating data. When the predicted residual life is lower than a preset safety threshold, the status monitoring and early warning module generates a fault early warning signal. S2. The operation and maintenance knowledge base construction and reasoning module receives the fault warning signal. The operation and maintenance knowledge base construction and reasoning module uses the power field knowledge graph to analyze the fault warning signal and generate an intelligent maintenance suggestion. The operation and maintenance knowledge base construction and reasoning module extracts a set of material lists from the intelligent maintenance suggestion. S3. The supply chain linkage decision module receives the material list set and the predicted residual life. The supply chain linkage decision module determines the supply response time of the warehouse node based on the global virtual inventory pool. The supply chain linkage decision module constructs a total cost objective function based on logistics costs and downtime losses. The supply chain linkage decision module solves the total cost objective function to determine the optimal supply source and generates a spare parts supply execution order. S4. The closed-loop feedback optimization module collects the true value data in the maintenance closure form. The closed-loop feedback optimization module uses the true value data to perform online correction operations on the time series data prediction model. The closed-loop feedback optimization module uses the true value data to perform dynamic update operations on the power field knowledge graph. The closed-loop feedback optimization module uses the true value data to perform adaptive calibration operations on the supply chain parameters.

2. The early warning and decision-making method for new energy equipment based on a multi-source knowledge base according to claim 1, characterized in that, In step S1, the specific steps for the condition monitoring and early warning module to calculate the predicted residual life of the new energy equipment include: The status monitoring and early warning module performs data cleaning processing on the real-time operating data; The status monitoring and early warning module maps the collected real-time operating data into equipment status monitoring vectors; The status monitoring and early warning module selects the device status monitoring vectors at multiple consecutive time points and constructs an input sequence matrix; The condition monitoring and early warning module inputs the input sequence matrix into the time series data prediction model to output the predicted residual lifetime.

3. The early warning and decision-making method for new energy equipment based on a multi-source knowledge base according to claim 1, characterized in that, In step S2, the specific steps for the operation and maintenance knowledge base construction and reasoning module to generate intelligent maintenance suggestions include: The operation and maintenance knowledge base construction and reasoning module uses retrieval enhancement generation technology to retrieve knowledge fragments in the power field knowledge graph that match the fault codes and abnormal feature vectors contained in the fault warning signal. The operation and maintenance knowledge base construction and reasoning module inputs the knowledge fragments and the fault warning signals into the large language model; The operation and maintenance knowledge base construction and reasoning module uses the large language model to perform reasoning and generation tasks to output the intelligent maintenance suggestion.

4. The early warning and decision-making method for new energy equipment based on a multi-source knowledge base according to claim 3, characterized in that, In step S2, the specific steps for the operation and maintenance knowledge base construction and reasoning module to construct the power field knowledge graph include: The operation and maintenance knowledge base construction and reasoning module extracts key entities from unstructured and semi-structured data. These key entities are divided into equipment objects, fault phenomena, fault causes, maintenance actions, required materials, and required tools. The operation and maintenance knowledge base construction and reasoning module defines the formal expression of the power domain knowledge graph, which consists of an entity set, a relation set, and a triplet set. The operation and maintenance knowledge base construction and reasoning module constructs a core fault handling link in the power field knowledge graph. The core fault handling link sequentially connects the fault code entity, the fault phenomenon entity, the component failure entity, the maintenance strategy entity, and the spare parts entity.

5. The method for early warning and decision-making of new energy equipment based on a multi-source knowledge base according to claim 1, characterized in that, In step S3, the specific steps by which the supply chain linkage decision-making module determines the supply response time of the warehouse node include: The supply chain linkage decision-making module constructs a global virtual inventory pool containing four inventory levels; The supply chain linkage decision module obtains the logistics time from the warehouse node to the site where the new energy equipment is located, as well as the procurement cycle corresponding to the material code. When the current available inventory of the warehouse node is greater than zero, the supply chain linkage decision module determines that the supply response time is equal to the logistics time. When the current available inventory of the warehouse node is zero, the supply chain linkage decision module determines that the supply response time is equal to the sum of the procurement cycle and the logistics time.

6. The early warning and decision-making method for new energy equipment based on a multi-source knowledge base according to claim 1, characterized in that, In step S3, the specific steps for the supply chain linkage decision-making module to construct the total cost objective function include: The supply chain linkage decision-making module defines the predicted residual lifetime as a safety window period. The supply chain linkage decision-making module obtains the logistics costs incurred from allocating materials from the warehouse node; The supply chain linkage decision-making module calculates the power generation loss cost generated per unit time when the new energy equipment is shut down. The supply chain linkage decision-making module calculates the unplanned downtime duration when the supply response time exceeds the safety window period; The supply chain linkage decision-making module defines the total cost objective function as the sum of the logistics cost and the downtime loss cost, wherein the downtime loss cost is equal to the power generation loss cost multiplied by the unplanned downtime duration.

7. The early warning and decision-making method for new energy equipment based on a multi-source knowledge base according to claim 6, characterized in that, In step S3, the specific steps for the supply chain linkage decision-making module to determine the optimal supply source include: The supply chain linkage decision-making module traverses all warehouse nodes in the global virtual inventory pool; The supply chain linkage decision-making module selects the warehouse node that minimizes the total cost objective function as the optimal supply source. When there is no available inventory in the global virtual inventory pool, the supply chain linkage decision module compares the safety window period with a preset emergency procurement threshold. If the safety window period is less than the emergency procurement threshold, the supply chain linkage decision module executes emergency mode. If the safety window period is greater than the emergency procurement threshold, the supply chain linkage decision module will execute the plan mode.

8. The early warning and decision-making method for new energy equipment based on a multi-source knowledge base according to claim 1, characterized in that, In step S4, the specific steps for the closed-loop feedback optimization module to collect true data include: The closed-loop feedback optimization module extracts the physical moment when the new energy equipment actually stops or fails, and defines the physical moment as the actual fault moment. The closed-loop feedback optimization module extracts the material codes and quantities of spare parts actually replaced by maintenance personnel on site, and defines the material codes and quantities of the actually replaced spare parts as actual material consumption data. The closed-loop feedback optimization module extracts the timestamp of the actual delivery of spare parts to the site where the new energy equipment is located. The closed-loop feedback optimization module defines the timestamp as the actual arrival time and obtains the order initiation time for generating the spare parts supply execution order.

9. The early warning and decision-making method for new energy equipment based on a multi-source knowledge base according to claim 8, characterized in that, In step S4, the specific steps for the closed-loop feedback optimization module to perform online correction include: The closed-loop feedback optimization module calculates the prediction deviation between the actual failure time and the corresponding time of the predicted residual lifetime. When the prediction deviation is greater than the allowable error threshold, the closed-loop feedback optimization module uses the corresponding input sequence matrix and the actual fault time to construct a corrected training sample pair. The closed-loop feedback optimization module uses the corrected training sample pair to perform incremental training operations on the time series data prediction model to adjust the network parameter set. Specifically, in step S4, the closed-loop feedback optimization module performs the dynamic update operation in the following steps: The closed-loop feedback optimization module locates the fault code entity corresponding to the real fault code and the spare parts entity corresponding to the actual material consumption data in the power field knowledge graph. When there is no material dependency relationship between the fault code entity and the spare parts entity, the closed-loop feedback optimization module creates a new triple in the power field knowledge graph; When there is already a dependent material relationship between the fault code entity and the spare parts entity, the closed-loop feedback optimization module uses a positive incentive algorithm to update the association weight value between the fault code entity and the spare parts entity. Specifically, in step S4, the closed-loop feedback optimization module performs the adaptive calibration operation of supply chain parameters, including the following steps: The closed-loop feedback optimization module calculates the actual delivery cycle based on the difference between the actual arrival time and the order initiation time. The closed-loop feedback optimization module updates the supplier delivery date correction coefficient using the actual delivery cycle and the pre-stored procurement cycle. The closed-loop feedback optimization module uses the updated supplier delivery time correction coefficient to dynamically correct the procurement cycle, so that the supply chain linkage decision module can recalculate the supply response time in subsequent steps.

10. A computer-readable storage medium, characterized in that, A computer-readable storage medium stores a computer program that, when executed by a processor, implements the new energy equipment early warning and decision-making method based on a multi-source knowledge base as described in any one of claims 1 to 9.

Images